Given that the last year saw the largest refinancing wave in the history of the single-family mortgage market, GNMA multifamily prepayment speeds over the past year have been fairly tame. This 12-month period covers April 2001 through March 2002, a period of historically low interest rates. Assuming a two-month time lag from rate-lock to closing for a project loan refinancing, the corresponding period for rate locks was February 2001 through January 2002, a period in which the 10-year Treasury yield ranged from 4.18 to 5.52 and averaged just 4.99. No production year except 1990 (which was likely inflated by the recent expiration of most prepayment penalties) had a one-year speed exceeding 24% CPR, and even 1990's 30.9% CPR was mild compared to premium single-family speeds. While there was some tendency of the highest coupon pools in the 1993-96 cohorts to prepay faster, overall prepayments were only modestly correlated with pool interest rate. The other motivation for refinancing, taking out equity, dominated, especially with the major increase in the value of real estate during the strong economy of the mid-to-late 1990's.
The last year's prepayment speeds give a good indication of how GNMA multifamily pools prepay in a relatively low interest rate environment. To get an indication of how they prepay in a much higher rate environment, we examine the one-year historical CPRs of the GNMA multifamily universe, broken down by production year and pass-through coupon, for the 12 months from July 1994 through June 1995. This period represents the interest rate peak of the past decade; the corresponding rate-lock period (assuming a 60-day lag) was May 1994 through April 1995, during which time the 10-year Treasury yield ranged from 6.91 to 8.03, and averaged 7.43. For that period, 1983 and later production had enough pools outstanding to generate a somewhat meaningful prepayment history.
In this high rate period, every production year except 1984 (a year represented by only 12 outstanding loans) from 1983 through 1990 prepaid at over 10% CPR. Production later than 1990 was mostly locked out at this time. So the indication from this relatively high rate year is that seasoned GNMA multifamily pools do not slow dramatically when interest rates rise. The desire of investors to take equity out of properties that have appreciated in value, whether for project expansion or improvements or to realize a cash return on the original investment, seems to prevent prepayment speeds from falling precipitately. This is in spite of the ability of project owners to extract equity for repairs, additions, or improvements with a section 241 second mortgage, or to recoup two years' worth of operating losses or out-of-pocket expenses with a section 221 (d) operating loss loan.
Analysis of historical CPRs over other recent periods can help us draw generalizations about prospective prepayment behavior for these GNMA pools. In Graph 1 below, we show historical one-year prepayments of the GNMA universe, broken down by production year, during the calendar years 1998 through 2001. For the corresponding 12-month rate-lock period (shifted two months earlier), the average 10-yr Treasury yields were 5.44 for 1998, 5.40 for 1999, 6.13 for 2000, and 5.11 for 2001. It appears that production years with recently expired call protection (i.e., in the 11th or 12th year since origination) tend to have prepayment spikes, after which their speeds slow considerably. These prepayment spikes occurred for 1987-88 production in 1998, for 1989 production in 1999, and for 1990-91 production in 2001. However, 1989-90 production exhibited no such rush to refinance in 2000, as the higher rates prevailing for most of that calendar year may have inhibited refinancing. The extensive refinancing of 1989 production during 1999 may have left few loans with a refinancing propensity remaining outstanding by 2000, while the higher rates of 2000 may have delayed some 1990 production refinancing until 2001, when that cohort saw a refinancing spike.
In summary, we think that the typical 15% CPR pricing assumption for voluntary prepayments on a diverse group of GNMA multifamily pools backing a REMIC is reasonable given this prepayment history, and that long-term speeds are unlikely to exceed 25% CPR in a low rate environment or fall much below 10% CPR in a high rate environment. Of course, this conclusion covers the entire GNMA multifamily universe; subsets of the universe containing relatively few loans can prepay very differently for loan-specific reasons. Investors can limit this idiosyncratic prepayment risk by diversifying their GNMA multifamily investments among several REMICs.
Defaults on GNMA pools
In the 1990s many of the REMICs backed by GNMA multifamilies were issued with pricing speeds that assumed no defaults would ever occur; i.e., that prior to lockout expiration, the collateral would experience no prepayments at all. Since the FHA was created in order to obtain financing for multifamily projects that would otherwise not be built at all (i.e., that the private sector would refuse to finance or would finance on such onerous terms that the project would not be economically viable), it would be very surprising if these securities experienced no defaults. In fact, we have identified 186 GNMA multifamily pools that have defaulted since the beginning of 1991, totaling $1.255 billion of original principal, and gaps in the GNMA multifamily database make it impossible to know whether we have identified every single default, although we think that we have identified most of them.
The default pattern over time is fairly clear and differs considerably from single-family mortgage default patterns, in which defaults are rare in the first two years after origination.
GNMA multifamily defaults occur with significant frequency in the very first year, rise in year 2, then level off through year 5, and decline fairly sharply after that. The early defaults probably reflect the fact that many FHA-insured projects serve primarily social policy goals, and some projects funded under FHA regulations would be considered fairly high credit risk loans by private lenders. A small fraction of these projects fail to become economically viable and default fairly soon after construction is completed. Those projects that generate sufficient income to service the mortgage during the first five years are likely to avoid default altogether.
We have identified the specific FHA insurance program covering 151 of these 186 defaulted loans (81.1% of the loans representing 82.5% of the $1.255 billion in defaulted GNMA multifamily principal). Section 223 (f) loans, those for purchase or refinancing of existing projects, had the worst default experience of any of the large programs. From the beginning of 1991 through March 2002, these loans, which comprise 16.6% of FHA loans and slightly over 20% of the original principal of the FHA universe, accounted for 35% of the GNMA multifamily defaulted loans and 42% of the GNMA multifamily defaulted principal. On the other hand, section 232 loans, which fund nursing homes and board and care homes, did not have a higher propensity to default than the FHA universe. Section 232 loans comprise 10.6% of FHA loans and 15.9% of the original principal of the FHA universe, and accounted for 12.9% of the GNMA multifamily defaulted loans but only 8.3% of the defaulted principal. Smaller FHA-insured nursing home projects have defaulted at somewhat higher rates than larger nursing home projects, but the overall experience of the CMBS market indicating that nursing home projects have more credit risk than apartment complexes does not appear to be true for the GNMA multifamily market.
The evidence from historical prepayment patterns is that the interest-sensitivity of GNMA multifamily pools is quite modest, certainly far less than that of GNMA single-family mortgages. In a low rate environment prepayments increase dramatically in the year after call protections expire, but then settle down to more normal CPRs in the teens or low 20s. In a high rate environment prepayments slow somewhat, but the declining value over time of tax advantages for project owners and the incentives to refinance to extract equity prevent long term speeds on seasoned pools with expired lockouts from falling much below 10% CPR. The explicit call protection on recently issued pools and the limited rate-sensitivity of refinancings on seasoned pools provides the investor with much better convexity than that provided by single-family MBS. Defaults, which are not interest-sensitive at all, have been falling somewhat in recent years but, given the nature of the loans, are unlikely to vanish from the GNMA multifamily market. Defaults create a little more cash flow uncertainty in the early years after issuance but do not create negative convexity because they are not significantly correlated with interest rates. The long-run prospects for default depend on a specific project's purpose and on underwriting policies and practices of the FHA, factors that have influenced them the most in the past.